Even with a strong context document and a well-constructed prompt, there will be sessions where the output is missing the mark. Before assuming the tool cannot produce what you need, run through this sequence.
Is the task instruction specific enough? Reread it. If it could be interpreted in more than one way, the tool has probably chosen the most generic interpretation. Narrow it.
Is there situational context you have not included? Something specific to this client, this moment, this task that the context document cannot know and the prompt has not provided?
Have you specified the format for this particular output? If the structure is wrong, the fix is almost always a format instruction rather than a different approach to the task.
Is there a constraint this specific task needs that your standing brief does not cover?
And have you written a follow-up prompt? One targeted follow-up, focused on the specific element that is not working, will resolve the vast majority of issues that survive the initial prompt.
If you have done all of that and the output is still not there, return to the task instruction and rebuild it from the beginning. Something in the brief is either missing or ambiguous. The tool is not being obstructive: it is working with what it has been given.
